awwwards-ui-skill

An open-source skill that raises the design quality of AI-generated web interfaces. When AI agents build UIs without guidance, they default to generic layouts, flat aesthetics, and missed accessibility standards. This skill fixes that — injecting design intent around motion, typography, color, and interaction directly into the agent's context.

Compatible with Antigravity, Claude Code, and Codex.

Examples

A developer portfolio — full-bleed hero, WebGL particle field, bento-box grid:

Portfolio demo

An agency site — asymmetric typography, glassmorphism overlays, mix-blend-mode cursor:

Agency demo

A product landing page — dark mode, 3D CSS object, scroll-triggered reveals:

Product landing demo

Installation

You can automatically install this skill to your Antigravity, Claude Code, and Codex config directories by running:

npx awwwards-ui-skill

If it prompts you to install the package, type y:

PS C:\Users\xyz> npx awwwards-ui-skill
Need to install the following packages:
[email protected]
Ok to proceed? (y) y

> npx
> awwwards-ui-skill

🚀 Installing Awwwards UI/UX Agent Skill...
✅ Successfully installed skill to C:\Users\xyz\.gemini\config\plugins\awwwards_ui_design\SKILL.md
✅ Successfully installed skill to C:\Users\xyz\.claude\skills\SKILL.md
✅ Successfully installed skill to C:\Users\xyz\.codex\skills\SKILL.md
🎉 You're all set! Tell your AI agent to build a 'cool ui' and watch the magic happen.

Manual Installation (Antigravity)

Clone into your plugins directory:

git clone https://github.com/Nyx-abu/awwwards-ui-skill.git ~/.gemini/config/plugins/awwwards-ui-skill

For Claude Code / Codex

Copy skills/awwwards_ui_design/SKILL.md directly into your custom system prompts or skill registry.

Usage

Just ask your AI assistant:

"Build me a portfolio with a cool UI"

The skill auto-triggers on keywords like cool ui, premium design, and awwwards style. The agent will apply structured design decisions — layout, motion, typography, color — and run an internal audit on accessibility and performance before returning any code.

Measurable outcome

The skill now requires a UI-Max Scorecard before an agent returns UI code. The target is 88/100 minimum, scored across:

• MOTION-01 easing, offset/delay, and fade paired with spatial movement
• MOTION-02 primary, secondary, and ambient motion layers
• LAYOUT-01 first-viewport composition and Awwwards-style hierarchy
• DEPTH-01 dimensionality through parallax, masking, morphing, or zoom continuity
• INTERACTION-01 microinteractions beyond color/opacity changes
• A11Y-01 semantic HTML, contrast, focus states, aria-hidden, and reduced motion
• PERF-01 requestAnimationFrame-safe pointer/scroll work and WebGL cleanup expectations
• RESP-01 responsive resilience at 375px, 768px, and 1440px
• BRAND-01 domain fit instead of generic visual spectacle

Run the deterministic verifier:

npm test

The test checks that the skill, README, handover, and sample website all include the scorecard contract and implementation evidence.

The sample site is at ../awwwards-ui-skill-test/index.html. It demonstrates the upgraded skill surface with a visible scorecard, custom cursor, magnetic controls, masked media reveal, parallax layer, zoom stage, bento grid, IntersectionObserver, prefers-reduced-motion, and requestAnimationFrame.

What It Covers

• Layouts: Full-bleed heroes, bento-box grids, asymmetric sections, horizontal scroll sequences
• Motion: GSAP ScrollTrigger reveals, Lenis smooth scrolling, parallax, text scramble transitions
• Cursors: RAF-driven custom cursors, magnetic snapping, mix-blend-mode: difference on hover
• 3D: Three.js particle fields and shader-based distortion tied to mouse input
• Color: Four curated palettes derived from top developer portfolio sites, with contrast rules baked in
• Typography: Display-scale contrast, tight heading tracking, variable font axis animation
• Accessibility: WCAG AA contrast enforcement, semantic HTML, ARIA roles, prefers-reduced-motion support
• Self-Audit: The agent checks its own output for layout, performance, and accessibility issues before returning code
• Measurement: The UI-Max Scorecard turns design quality into auditable pass/fail evidence instead of subjective taste language

Contributing

See CONTRIBUTING.md for full guidelines. The short version:

1. Fork the repo
2. Create a branch: git checkout -b feat/your-idea
3. Make focused changes to SKILL.md or add new patterns
4. Open a PR with a description and an example of the generated output

We are currently looking for testers to spot repeated AI pattern bugs before the v1 stable release. Details in CONTRIBUTING.md.

License

MIT — free to use, fork, and build on.